Modeling physical properties of chemical mixtures is an important task in many industries and processes. Specifically, for many processes, accurate modeling of physical properties for various mixtures is crucial for such areas as process design and process control applications. For example, modeling physical properties of chemical mixtures is often useful when selecting suitable solvents for use in chemical processes.
Solvent selection is an important task in the chemical synthesis and recipe development phase of the pharmaceutical and agricultural chemical industries. The choice of solvent can have a direct impact on reaction rates, extraction efficiency, crystallization yield and productivity, etc. Improved solvent selection brings benefits, such as faster product separation and purification, reduced solvent emission and less waste, lower overall costs, and improved production processes.
In choosing a solvent, various phase behavior characteristics of the solvent-solute mixtures are considered. For example, vapor-liquid equilibrium (VLE) behavior is important when accounting for the emission of solvent from reaction mixtures, and liquid-liquid miscibility (LLE) is important when a second solvent is used to extract target molecules from the reaction media. For solubility calculations, solid-liquid equilibrium (SLE) is a key property when product isolation is done through crystallization at reduced temperature or with the addition of anti-solvent.
For many applications, hundreds of typical solvents, not to mention an almost infinite number of mixtures thereof, are candidates in the solvent selection process. In most cases, there is simply insufficient phase equilibrium data on which to make an informed solvent selection. For example, in pharmaceutical applications, it is often the case that phase equilibrium data involving new drug molecules in the solvents simply do not exist. Although limited solubility experiments may be taken as part of the trial and error process, solvent selection is largely dictated by researchers' preferences or prior experiences.
Many solubility estimation techniques have been used to model the solubility of components in chemical mixtures. Some examples include the Hansen model and the UNIFAC group contribution model. Unfortunately, these models are rather inadequate because they have been developed mainly for petrochemicals with molecular weights in the 10 s and the low 100 s daltons. These models do not extrapolate well for chemicals with larger molecular weights, such as those encountered in pharmaceutical applications. Pharmaceuticals are mostly large, complex molecules with molecular weight in the range of about 200-600 daltons.
Perhaps, the most commonly used methods in solvent selection process are the solubility parameter models, i.e., the regular solution theory and the Hansen solubility parameter model. There are no binary parameters in these solubility parameter models and they all follow merely an empirical guide of “like dissolves like.” The regular solution model is applicable to nonpolar solutions only, but not for solutions where polar or hydrogen-bonding interactions are significant. The Hansen model extends the solubility parameter concept in terms of three partial solubility parameters to better account for polar and hydrogen-bonding effects.
In his book, Hansen published the solubility parameters for over 800 solvents. See Hansen, C. M., HANSEN, SOLUBILITY PARAMETERS: A USER'S HANDBOOK (2000). Since Hansen's book contains the parameters for most common solvents, the issue in using the Hansen model lies in the determination of the Hansen solubility parameters from regression of available solubility data for the solute of interest in the solvent selection process. Once determined, these Hansen parameters provide a basis for calculating activity coefficients and solubilities for the solute in all the other solvents in the database. For pharmaceutical process design, Bakken, et al. reported that the Hansen model can only correlate solubility data with ±200% in accuracy, and it offers little predictive capability. See Bakken, et al., Solubility Modeling in Pharmaceutical Process Design, paper presented at AspenTech User Group Meeting, New Orleans, La., Oct. 5-8, 2003, and Paris, France, Oct. 19-22, 2003.
When there are no data available, the UNIFAC functional group contribution method is sometimes used for solvent selection. In comparison to the solubility parameter models, UNIFAC's strength comes with its molecular thermodynamic foundation. It describes liquid phase nonideality of a mixture with the concept of functional groups. All molecules in the mixture are characterized with a set of pre-defined UNIFAC functional groups. The liquid phase nonideality is the result of the physical interactions between these functional groups and activity coefficients of molecules are derived from those of functional groups, i.e., functional group additivity rule. These physical interactions have been pre-determined from available phase equilibrium data of systems containing these functional groups. UNIFAC gives adequate phase equilibrium (VLE, LLE and SLE) predictions for mixtures with small nonelectrolyte molecules as long as these molecules are composed of the pre-defined set of functional groups or similar groups.
UNIFAC fails for systems with large complex molecules for which either the functional group additivity rule becomes invalid or due to undefined UNIFAC functional groups. UNIFAC is also not applicable to ionic species, an important issue for pharmaceutical processes. Another drawback with UNIFAC is that, even when valuable data become available, UNIFAC cannot be used to correlate the data. For pharmaceutical process design, Bakken et al., reported that the UNIFAC model only predicts solubilities with a RMS (root mean square) error on ln x of 2, or about ±500% in accuracy, and it offers little practical value. Id.
A need exists for new, simple, and practical methods of accurately modeling one or more physical properties of a mixture.